The continuous “Moore's Law” scaling of semiconductors to higher densities and greater performance has provided tremendous increases in productivity for the industry and our society. However, a problem resulting from this scaling is the requirement to carry ever-higher electrical currents in smaller and smaller interconnection wires. When the current densities and temperatures in such small wires get too high, the interconnect wires can fail by a phenomenon called electromigration. The effects of the so-called “electron wind” that occurs in high current-density interconnect wires causes metal atoms to be swept away from their original lattice positions, resulting in either an open circuit in the wire or an extrusion short in an area where these diffusing metal atoms collect. The introduction of copper as a wiring material to replace aluminum provided a tremendous improvement in electromigration lifetime, however the continued scaling of interconnect wiring suggests that additional improvements in copper electromigration lifetime will be needed in the future.
Unlike aluminum interconnects, which fail by diffusion of aluminum atoms along grain boundaries, the copper interconnect electromigration failure mode is controlled by diffusion along surfaces and interfaces. In particular, for conventional copper wiring interconnect schemes, the top surface of a copper wire typically has an overlying dielectric capping layer, which must have good diffusion barrier properties to prevent migration of copper into the surrounding dielectric. The two most commonly used dielectric capping materials are silicon nitride and silicon carbon nitride, which are conventionally deposited by a plasma-enhanced chemical vapor deposition (PECVD) technique. Unfortunately, these PECVD deposited capping materials form a defective interface with the copper that results in enhanced copper migration along the top surface of the copper wire and therefore higher electromigration failure rates. Other surfaces of a copper wire structure typically have interfaces with a barrier layer or bi-layer (typically metallic, as for example TaN/Ta, TaN/Ru, or Ru) that forms a strong interface with the copper to limit copper diffusion and therefore suppress electromigration effects. We refer to such a barrier layer or bi-layer as a “barrier layer”.
There have been attempts to improve electromigration of copper wiring by capping the top surfaces of copper wires with selectively deposited metal caps. Indeed, when the top copper interface has been capped with either selective tungsten or a selective cobalt tungsten phosphide (CoWP) metallic layer, tremendous improvements in copper electromigration lifetimes have been reported. Unfortunately, all of the methods that use a selective metallic capping solution have some probability of also depositing some metal on the adjacent insulator surface and therefore causing unintended leakage or shorts between adjacent metal lines. The present invention uses gas-cluster ion-beam processing to solve many of these problems.
FIG. 1 shows a schematic illustrating the wiring scheme 300 of a prior art silicon nitride capped copper interconnect, as commonly used in the copper dual damascene integration process. It comprises a first copper wire layer 302, a second copper wire layer 304, and copper via structure 306 connecting the two copper layers. Sidewalls and bottoms of both wire layers 302 and 304 and the via structure 306 are all lined with a barrier layer 312. The barrier layer 312 provides excellent diffusion barrier properties, which prevent diffusion of the copper into the adjacent insulator structure and also provides an excellent low diffusion interface with the copper that suppresses electromigration along these interfaces. First inter-level dielectric layer 308 and second inter-level dielectric layer 310 provide insulation between the copper wires. The top surface of first copper wire layer 302 and the top surface of the second copper wire layer 304 are each covered with insulating barrier films 314 and 316, respectively, which are typically composed of silicon nitride or silicon carbon nitride. These insulating barrier films 314 and 316 are conventionally deposited by PECVD and the interfaces that they form with the exposed copper surfaces are rather defective and offer fast diffusion paths for migrating copper atoms. In this prior art wiring scheme, it is along these interfaces that almost all of the undesirable material movement occurs during copper electromigration. In conventional dual damascene copper interconnects like this, at each interconnect level, after formation of trenches and vias in the inter-level dielectric layer and subsequent deposition of copper to form interconnect wires and vias, there follows a planarization step typically performed using chemical mechanical polishing (CMP) techniques. Corrosion inhibitors are used in both CMP and post-CMP brush cleaning processes and these corrosion inhibitors and other contaminants must be removed from the copper surface with an in-situ cleaning prior to the deposition of a capping layer. Use of an ex-situ cleaning process would leave the copper surface vulnerable to corrosion and oxidation. PECVD reactors are typically not configured to perform an effective in-situ cleaning of the copper surface prior to the insulator capping layer deposition. Although not shown in FIG. 1, the wiring scheme 300 is typically formed on a semiconductor substrate containing active and/or passive elements requiring electrical interconnection to complete an integrated circuit.
FIG. 2 shows a wiring scheme 400 of a prior art selective metal-capped copper interconnect. It comprises a first copper wire layer 402, a second copper wire layer 404, and copper via structure 406 connecting the two copper layers. The sidewalls and bottoms of both wire layers 402 and 404 and the via structure 406 are all lined with a barrier layer 412. The barrier layer 412 provides excellent diffusion barrier properties, which prevent diffusion of the copper into the adjacent insulator structure and also provides an excellent low diffusion interface with the copper that suppresses electromigration along these interfaces. First inter-level dielectric layer 408 and second inter-level dielectric layer 410 provide insulation between copper wires. The top surface of first copper wire layer 402 and the top surface of the second copper wire layer 404 are each capped with selectively deposited metallic layers 414 and 416, respectively, which are typically composed of either selective tungsten or selective CoWP deposited by chemical vapor deposition (CVD) or an electroless technique, respectively. In this conventional dual damascene copper interconnect, at each interconnect level, after formation of trenches and vias in the inter-level dielectric layer and subsequent deposition of copper to form interconnect wires and vias, there follows a planarization step typically performed using chemical mechanical polishing (CMP) techniques. Corrosion inhibitors are used in both CMP and post-CMP brush cleaning processes and these and other contaminants must be removed from the copper surface prior to the capping layer deposition. When the top copper interface of a copper layer has been capped with either a tungsten or a CoWP metallic layer, large improvements in copper electromigration lifetimes have been reported. Unfortunately, all of the methods that use a selective metallic capping solution have some probability of also depositing unwanted metal 418, shown for example, on adjacent insulator surfaces, and therefore can result in electrical leakage or shorts between adjacent metal lines. Although selective metal deposition techniques offer the promise of very large electromigration improvements, they have not been widely implemented in manufacturing because of the high potential for loss of yield on semiconductor die due to the deposition of unwanted metal. Although not shown in FIG. 2, the wiring scheme 400 is typically formed on a semiconductor substrate containing active and/or passive elements requiring electrical interconnection to complete an integrated circuit.
The use of a gas-cluster ion beam for processing surfaces is known (see for example, U.S. Pat. No. 5,814,194, Deguchi et al.) in the art. As the term is used herein, gas-clusters are nano-sized aggregates of materials that are gaseous under conditions of standard temperature and pressure. Such gas-clusters typically consist of aggregates of from a few to several thousand molecules loosely bound to form the gas-cluster. The gas-clusters can be ionized by electron bombardment or other means, permitting them to be formed into directed beams of controllable energy. Such ions each typically carry positive charges of q·e (where e is the electronic charge and q is an integer of from one to several representing the charge state of the gas-cluster ion). Non-ionized gas-clusters may also exist within a gas-cluster ion beam. The larger sized gas-cluster ions are often the most useful because of their ability to carry substantial energy per gas-cluster ion, while yet having only modest energy per molecule. The gas-clusters disintegrate on impact, with each individual molecule carrying only a small fraction of the total gas-cluster ion energy. Consequently, the impact effects of large gas-cluster ions are substantial, but are limited to a very shallow surface region. This makes gas-cluster ions effective for a variety of surface modification processes, without the tendency to produce deeper subsurface damage characteristic of conventional monomer ion beam processing. Means for creation of and acceleration of such GCIBs are described in the reference (U.S. Pat. No. 5,814,194) previously cited. Presently available gas-cluster ion sources produce gas-cluster ions having a wide distribution of sizes, N (where N=the number of molecules in each gas-cluster ion—in the case of monatomic gases like argon, an atom of the monatomic gas will be referred to as a molecule and an ionized atom of such a monatomic gas will be referred to as a molecular ion—or simply a monomer ion—throughout this discussion). Many useful surface-processing effects can be achieved by bombarding surfaces with GCIBs. These processing effects include, but are not necessarily limited to, cleaning, smoothing, etching, doping, and film formation or growth. U.S. Pat. No. 6,537,606, Allen et al., teaches the use of GCIBs for corrective etching of an initially non-uniform thin film to improve its spatial uniformity. The entire content of U.S. Pat. No. 6,537,606 is incorporated herein by reference.
Upon impact of an energetic gas-cluster on the surface of a solid target, penetration of the atoms of the cluster into the target surface is typically very shallow because the penetration depth is limited by the low energy of each individual constituent atom and depends principally on a transient thermal effect that occurs during the gas-cluster ion impact. Gas-clusters dissociate upon impact and the individual gas atoms then become free to recoil and possibly escape from the surface of the target. Other than energy carried away by the escaping individual gas atoms, the total energy of the energetic cluster prior to impact becomes deposited into the impact zone on the target surface. The dimensions of a target impact zone are dependent on the energy of the cluster but are on the order of the cross-sectional dimensions of the impacting cluster and are small, for example, roughly 30 Angstroms in diameter for a cluster comprised of 1000 atoms. Because of the deposition of most of the total energy carried by the cluster into the small impact zone on the target, an intense thermal transient occurs within the target material at the impact site. The thermal transient dissipates quickly as energy is lost from the impact zone by conduction deeper into the target. Duration of the thermal transient is determined by the conductivity of the target material but will typically be less than 10−6 second.
Near a gas-cluster impact site, a volume of the target surface can momentarily reach temperatures of many hundreds to several thousands of degrees Kelvin. As an example, impact of a gas-cluster carrying 10 keV total energy has been estimated to be capable of producing a momentary temperature increase of about 2000 degrees Kelvin throughout a highly agitated, approximately hemispherical zone extending to about 100 Angstroms below the surface. This high thermal transient promotes intermixing and reaction of the workpiece and gas-cluster ion beam constituents and thus results in improved electromigration lifetime.
Following initiation of an elevated temperature transient within the target volume below an energetic gas-cluster impact site, the affected zone cools rapidly. Some of the gas-cluster constituents escape during this process, while others remain behind and become incorporated in the surface. A portion of the original surface material may also be removed by sputtering or like effects. In general, the more volatile and inert constituents of the gas-cluster are more likely to escape, while the less volatile and more chemically reactive constituents are more likely to become incorporated into the surface. Although the actual process is likely much more complex, it is convenient to think of the gas-cluster impact site and the surrounded affected zone as a “melt zone” wherein the gas-cluster atoms may briefly interact and mix with the substrate surface and wherein the gas-cluster materials either escape the surface or become infused into the surface to the depth of the affected zone. The term “infusion” or “infusing” is used by the inventors to refer to this process to distinguish it from ion “implantation” or “implanting”, a very different process that produces very different results. Noble gases in the energetic gas-cluster ion, such as argon and xenon, for example, being volatile and non-reactive have a high probability of escape from the affected zone, while materials such as carbon, boron, fluorine, sulfur, nitrogen, oxygen, germanium, and silicon, for example, being less volatile and/or more likely to form chemical bonds, are more likely to remain in the affected zone, becoming incorporated in the surface of the substrate.
Noble inert gases such as argon and xenon, for example, not for limitation, can be mixed with gases containing elements that are less volatile and more reactive to form mixed clusters. Such gas-clusters can be formed with existing gas-cluster ion beam processing equipment as will be described hereinafter, by using suitable source gas mixtures as the source gas for gas-cluster ion beam generation, or by feeding two or more gases (or gas mixtures) into the gas-cluster ion generating source and allowing them to mix in the source. In a recent publication, Borland et al. (“USJ and strained-Si formation using infusion doping and deposition”, Solid State Technology, May 2004, p 53) have shown that GCIB infusion can produce graded surface layers transitioning smoothly from the substrate material to the deposited layer on the surface.